I am probably not in the majority in finding this wholly unsurprising:
THE GOVERNMENT revealed Thursday that a Tesla Model S crashed into a truck in Florida in May, killing the electric car’s driver.
In the Florida case, the car failed to detect a large truck that had crossed into the Tesla’s path, perhaps because it blended in with a brightly lit sky.
A lot of people who are not engineers, and a lot of people who are, have a touching faith in the ability of technology to solve every problem there is. People tend to look at technological progress in one area and assume that it can be seamlessly transposed into something entirely different provided enough minds are working on it. This is why people are so optimistic about self-driving cars: they look at the amazing advances in computer power in the past few decades, they see Google has a huge stack of money and a very clever search engine, and conclude that self-driving cars are just a matter of time because…well, technology, innit? And the same people often think it is self-evident that computers will always do a better job than humans as the former are infallible.
Personally, I understand enough about how things work to know that technological progress in any given area is not inevitable, there must be a mechanism in place for the shift to happen, e.g. a step-change in technology in the vein of the PCB or transistor. When you consider how long the humble motor car has been around and the hundreds of millions of manhours that have been spent in trying to improve it in every possible way, it is astonishing how little has changed since the Model T Ford. The basic principles of how a car is powered, controlled, and physically laid out haven’t changed. They even still have wing mirrors and a driver’s rear view mirror. So much for technology. There have been plenty of improvements and enhancements, but no step-change in motor car technology since the first one rolled off a mass production line.
Google reckon they can make his step-change by doing away with the driver, and everyone seems to be confident they, or somebody else, will be successful in doing so. Why I’m don’t share their confidence is because of two technical reasons: the first, which I’ll write about at length in a separate post, is the cost of manufacturing, testing, and maintaining extremely reliable electronic systems. The second is that I do not believe computers will ever be as good as humans at driving in the environment in which humans live.
The mistake people make is to assume every action in driving is one of simple measurement, and conclude that computers are far better at measuring things than humans are in terms of speed and accuracy. However, driving is often about judgement as opposed to pure measurement (and this is why it takes a while to become a good driver, judgement improves with experience), and much of this judgement relates to the interpretation of visual information. The recognition of objects by computers is still only in its infancy, and nowhere near robust enough to deploy in any safety-critical system. Given the pace of development of other areas of computing abilities, such as sound recognition in apps like Shazam, object recognition is seriously lagging behind and I suspect for very good reasons: software, being made up of pre-programmed algorithms, simply isn’t very good at it. And even then object recognition isn’t enough, a self-driving car would need to be able to not only accurately acquire visual data but also interpret it before initiating an action (or not). Computers are unable to do this for anything other than the most basic of pre-determined objects and scenarios, while the environment in which humans operate their cars is fiendishly complex.
There are those who think that advances in computing power will solve this issue, but I think the problem of visual data acquisition and interpetation is one more akin to aesthetics than measurement, i.e. its a judgement, not a binary decision. Are we confident a computer will one day be able to write a decent novel? Or generate a picture which is not a pre-programmed mathematical model which the coder knew in advance produces nice shapes? With enough computing power, do we believe a computer could write a better song than a human could? Personally, I don’t think this will ever happen because so much of aesthetics is down to judgement and involves variables which cannot be properly defined, much less defined in advance in a piece of code.
I believe a human’s ability to determine at a glance that an object in the road is a shallow puddle and not a large rock is the same ability which can differentiate between an operatic aria and a pet shop on fire. Computers don’t have this ability, as the failure of Tesla’s to tell the difference between a large truck and the sky shows. What does amaze me though is that computers are being put into cars with the belief that they can do things they demonstably can’t. A hefty lawsuit and tighter regulations can’t be too far away.
If self-driven cars have a future, I believe they will take the form of manually-controlled machines which switch to self-drive mode only once they are driven by a human onto a very tightly controlled and sterilised environment such as a motorway specifically designed to take only self-driving vehicles. I am confident we will never see self-driving vehicles moving around cities and towns as we currently know them, ever.